arXiv:2611.21651v1 Announce Type: new Abstract: A controller that resamples an answer, escalates to a larger model, or abstains is usually justified by a confidence signal. Three claims about such a signal do not follow from one another: it ranks answers (discrimination), it supports withholding some of them (selective prediction), and it changes an answer for the better (correction utility). We give an exact decomposition of the third into gated recoveries minus gated harms, which bounds whether a comparison can succeed before it is run: a gate that only replaces a proposal can act only where the base policy was already wrong, so the initial error count caps the discordant pairs any correct gate can produce. When that cap falls below the count a paired test needs, the experiment cannot return a positive result for a working mechanism, nor distinguish one from a gate that never fired. We measure all three targets on two model scales and find the bound binds. In a small-model suite a perturbation-agreement verifier reaches within-policy AUROC 0.990 (95% interval [0.95, 1.00]) while every episode outcome matches single-sample decoding at 4.32 times the call count. In a controlled 840-episode study no verifier replaced the proposal on any of the 60 GSM8K decisions, so that comparison tested nothing; the entire GSM8K gain across six controllers is one task, and it is the task whose identical temperature-zero repeats disagreed, making it the one answer the service does not reproduce. On the diagnostic set the gate could reach four errors while the test needed six, so even a perfect gate had no significant result available. We report two checks that would have caught this before any budget was spent, and show that tie handling in rankings, tied selection cutoffs, and unequal output limits each move a headline number in the same dataset. The binding constraint on this literature is measurement resolution, not the choice of signal.
Consistency Is Not Correction: Separating Confidence, Abstention, and Recovery in LLM Agents
A study on arXiv (2611.21651v1) finds that confidence signals in LLM agent controllers can rank answers well without enabling correction, because a gate that only replaces a proposal can act only where the base policy was already wrong, capping the discordant pairs any correct gate can produce. In a small-model suite a perturbation-agreement verifier reached within-policy AUROC 0.990 (95% interval [0.95, 1.00]) while every episode outcome matched single-sample decoding at 4.32 times the call count, and in a controlled 840-episode study no verifier replaced the proposal on any of the 60 GSM8K decisions, so that comparison tested nothing. The authors report that the entire GSM8K gain across six controllers is one task — the one whose identical temperature-zero repeats disagreed — and that on the diagnostic set the gate could reach four errors while the test needed six, concluding the binding constraint on this literature is measurement resolution, not the choice of signal.
Before running a controller comparison, compute the initial error count as the cap on discordant pairs and confirm it exceeds what your paired test needs; otherwise the experiment cannot distinguish a working gate from one that never fired.
more in playbook: Evals and testing for LLM appsRun your AI side-project on zahid.host
EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.